By name: colab-debug
description: Debugging known Google Colab runtime issues — install failures, segfaults, kernel crashes, shell quirks. Invoke when the user reports an error or unexpected behavior in an existing Colab notebook. For interactively writing/editing a Colab notebook via the colab-mcp bridge, use the colab-mcp skill instead.
argument-hint: [topic]
Google Colab Debugging Reference
For interactive editing of a Colab notebook (Claude adding/running cells in a live browser
session), see the colab-mcp skill. This skill is the catalog of known fixes for problems
that come up inside a Colab runtime.
Debugging workflow
Gather diagnostics in the current session before restarting — restarting loses all installed packages and downloaded files. Only suggest notebook edits once the fix is confirmed from diagnostics. Then tell the user "good time to restart and apply edits" so they restart once with the correct fix rather than iterating through multiple restarts.
Known issues and fixes
wgetsilent failures: always use-qfor progress suppression but check file sizes after download (!ls -lh). 0-byte files mean the URL was inaccessible (e.g. university servers block non-campus IPs). Use Google Drive +gdowninstead.blastnsegfault (SIGSEGV): usually a version mismatch between the binary and the database. The apt package (ncbi-blast+) installs 2.12.0; install the correct version from NCBI FTP and overwrite withcp ... $(which blastn).pysaminstall failure: needs system headers before pip can compile it.apt-get install libbz2-dev liblzma-dev libcurl4-openssl-devfirst, thenpip install pysam.condacolab: intentionally crashes the kernel to restart — shows as an error in the VSCode Colab extension. Avoid; prefer NCBI binaries or pip where possible.Multiline
!commands with\: unreliable in Colab. Use%%bashcell magic for multiline shell commands instead.